In this study, we have identified genes expressed in SAT and VAT that are related to lipid and glucose metabolism parameters in obesity. In particular, plasma levels of HDL cholesterol and glucose were found to be correlated to sets of co-expressed, and thus functionally related, genes (modules). Remarkably, several SAT modules were correlated to plasma HDL cholesterol levels and one VAT module was correlated to plasma glucose levels, although these SAT modules contained primarily the same genes as the VAT module. This difference highlights the fact that SAT and VAT have a distinct biological role. In silico classification of the co-expressed genes revealed that a significant number are involved in immunity and metabolism. This is in line with the concept that the immune and metabolic systems are tightly interconnected and that this interconnection is pivotal in the development of co-morbidities of obesity.
Several of the genes we have identified in this study play a role in pathways or processes that have already been linked to obesity co-morbidity, in particular HDL levels. These pathways or processes include immunity-related signalling pathways, the complement cascade, cholesterol metabolism and trafficking, lysosomal degradation and trafficking, and composition of the HDL particle [30–40]. A crucial finding of this study is the identification of novel genes that are correlated to HDL and glucose levels in severely obese individuals. The role of these genes in obesity co-morbidity is largely unknown, and further research is required to unravel the relationship between these genes and HDL and glucose levels. Possibly these genes may control plasma HDL cholesterol and glucose levels, or they might be involved in the response of adipose tissue to changed plasma HDL and glucose levels.
An earlier micro array study performed by Tchkonia et al  investigated differences in gene expression levels between differentiated and undifferentiated adipocytes derived from subcutaneous, visceral, and mesenteric adipose tissue. We observed an overlap between the results of this study and our own data that was higher than expected: In our study we identified 1344 genes to be upregulated in SAT and 1246 in VAT. Of these 1344 and 1246 genes, 103 and 87 respectively had also been identified in the study of Tchkonia et al, which identified 920 transcripts to be differently expressed across fat depots in either differentiated or undifferentiated cells (overrepresentation p-values: 2.5 × 10-7 for SAT and 8.7 × 10-5 for VAT, Fisher's exact test assuming that 20,000 unique genes were tested in total). Of the 87 genes differentially expressed in the study of Tchkonia et al that overlapped with genes upregulated in VAT in our study, 76 (87%) were differentially expressed in undifferentiated adipocytes that had been derived from distinct fat depots. 39 of these 76 genes (51%) were present in module VAT 4, which is substantially higher than expected (p-value: 2.5 × 10-11; Fisher's exact test, again assuming that 20,000 different genes were tested). These observations make it tempting to speculate that this module is related to processes in VAT-specific undifferentiated adipocytes. This is line with absence of a module in SAT that contains the same genes as module VAT 4.
Previous studies on the effects of obesity on genome-wide expression levels in SAT revealed several classes of genes to be regulated by obesity [12, 13]. Downregulated genes in obesity include lipolytic genes. Upregulated genes include genes controlling the structure and turnover of the extracellular matrix (ECM) and genes of infiltrating immune cells encoding cytokines and plasma membrane proteins. Another study investigating gene expression levels in whole SAT before and after weight loss  found similar sets of genes as found in the studies mentioned above [12, 13]. A subset of these genes was shown to be linked to glucose disposal rate, indicating that they may be involved in insulin resistance. Among the genes involved in immunity and the ECM, there was an overrepresentation of genes expressed in immune cells (e.g. macrophages), whereas genes involved in lipid metabolism were mostly genes expressed in adipocytes. Investigation of 31 genes specifically expressed in macrophages but not in adipocytes  revealed that these genes show significantly different gene expression profiles during weight loss induced by a stringent diet. 2 genes did not respond to this diet, whereas 7 genes responded strongly, 11 genes responded weakly, and another 11 genes showed an intermediate response .
In our studies of a group of 75 severely obese individuals, the genes in SAT modules 4 and 8, and VAT module 9 showed significant overlap with the genes differentially expressed after weight loss , as well as with the genes differentially expressed between lean and obese individuals  (Additional file 13, Table S10A). In addition, genes in these modules overlap with macrophage genes differentially expressed during dietary intervention and with genes predictive of insulin sensitivity (Additional file 13, Table S10B and S10C). Strikingly, of the 31 genes specifically expressed in macrophages but not in adipocytes investigated by Capel et al. , 26 are present in SAT modules 4 and 8, which are correlated to HDL levels (p = 2.1 × 10-40; Fisher's exact test assuming that 20,000 different genes were tested). Moreover the grouping of these macrophage genes based on different expression patterns during dietary intervention closely resembles the grouping of the genes in SAT modules 4 and 8 generated in our study; 6 of 7 genes identified as high responders to energy restriction regarding their expression are present in module SAT 4, and 10 of 11 genes identified as low responders are present in module SAT 8 . The overlap of genes found in these studies with different designs - comparing lean with obese individuals, studying the same individuals after weight loss, and studying quantitative metabolic traits in obese individuals - supports these approaches and strongly suggests that the genes identified are involved in obesity-related disease mechanisms.
It should be noted that the correlations between the modules and the metabolic traits identified in our study are not driven by BMI, since BMI itself was not correlated to the modules - except VAT 40. The correlation between module SAT 8 and plasma HDL levels was confounded by BMI and plasma insulin levels, but the absence of any correlation between this module and BMI or plasma insulin levels after correction for plasma HDL levels, indicates that HDL is the driver of this module. It can be speculated that module SAT 8 represents a BMI/plasma insulin driven effect of HDL whereas module SAT 4 represents an effect of HDL independent of BMI/plasma insulin.
A remaining question is what biological phenomena are driving the modules correlated to a metabolic trait. Here, we will mainly focus on SAT modules 4 and 8, because these two modules contain the largest number of genes, making it more valid to identify over-represented pathways in them. Capel et al.  investigated 31 genes specifically expressed in macrophages but not in adipocytes. Of these, 26 are present in SAT modules 4 and 8, which are correlated to HDL levels. The presence of genes within these modules that are specifically expressed in macrophages, might be a reflection of the relative number of macrophages in the whole adipose tissue and it is possible that SAT modules 4 and 8 are, at least in part, driven by the degree of macrophage infiltration. The presence of two different modules of macrophage genes, as confirmed by grouping of the macrophage genes by Capel et al., might be driven by differences between or within macrophages (e.g. macrophage infiltration or activation). Another possible biological mechanism that might underlie the appearance of SAT modules 4 and 8 are differences in adipocyte size, since with equal numbers of macrophages per m3 the relative amount of macrophage mRNA would increase if adipocytes get larger. Other mechanisms that may operate are the induction of adipocyte autophagy, ER-stress, or inflammasome activation. In order to get insight into these questions histology experiments are required to quantitate macrophage infiltration, adipocyte size, markers of autophagy, ER-stress, and inflammasome activation. Of note, 13 of the 31 macrophage-specific genes were also present in VAT module 9. This overlap is still highly significant, although less striking than in SAT (p = 1.6 × 10-22; Fisher's exact test assuming that 20,000 different genes were tested).
Importantly, VAT is the most metabolically active fat depot, and it has been proposed that complications of obesity correlate to an excess of visceral fat rather than to subcutaneous fat accumulation [9, 10]. However, many studies investigating gene expression in adipose tissue have only focused on SAT. In our study, we included samples from both fat depots and we identified numerous genes differentially expressed in VAT and SAT. In this regard our results are in line with a previous study in ten nondiabetic, normolipidemic obese men , but due to our larger sample size we were able to detect more genes differentially expressed in SAT and VAT.
Unexpectedly the module correlated to plasma glucose levels in VAT contained primarily the same genes as the modules correlated to plasma HDL cholesterol levels in SAT. Moreover the genes differentially expressed in SAT and VAT were not correlated to any of the parameters we tested, and expression levels of the genes that were correlated to plasma HDL levels in SAT and glucose levels in VAT were similar in both tissue types. This could indicate that although gene expression levels in VAT and SAT are associated with different plasma parameters - glucose and HDL levels, respectively - the molecular perturbations that underlie these associations are the same. Further, it might imply that gene expression in SAT is a reasonably good "model" for gene expression in VAT in regard to HDL and glucose metabolism.